caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
fc712eb7aa00aabcdafda54776038efdc486d570
clang-p2996/mlir/lib/Conversion/SCFToStandard/SCFToStandard.cpp
Alex Zinenko 1ec60862d7 [mlir] Avoid cloning ops in SCF parallel conversion to CFG 
The existing implementation of the conversion from SCF Parallel operation to
SCF "for" loops in order to further convert those loops to branch-based CFG has
been cloning the loop and reduction body operations into the new loop because
ConversionPatternRewriter was missing support for moving blocks while replacing
their arguments. This functionality now available, use it to implement the
conversion and avoid cloning operations, which may lead to doubling of the IR
size during the conversion.

In addition, this fixes an issue with converting nested SCF "if" conditionals
present in "parallel" operations that would cause the conversion infrastructure
to stop because of the repeated application of the pattern converting "newly"
created "if"s (which were in fact just moved). Arguably, this should be fixed
at the infrastructure level and this fix is a workaround.

Reviewed By: herhut

Differential Revision: https://reviews.llvm.org/D91955
2020-11-23 14:01:22 +01:00

594 lines
26 KiB
C++
Raw Blame History

//===- SCFToStandard.cpp - ControlFlow to CFG conversion ------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements a pass to convert scf.for, scf.if and loop.terminator
// ops into standard CFG ops.
//
//===----------------------------------------------------------------------===//
#include "mlir/Conversion/SCFToStandard/SCFToStandard.h"
#include "../PassDetail.h"
#include "mlir/Dialect/SCF/SCF.h"
#include "mlir/Dialect/StandardOps/IR/Ops.h"
#include "mlir/IR/BlockAndValueMapping.h"
#include "mlir/IR/Builders.h"
#include "mlir/IR/BuiltinOps.h"
#include "mlir/IR/MLIRContext.h"
#include "mlir/IR/PatternMatch.h"
#include "mlir/Transforms/DialectConversion.h"
#include "mlir/Transforms/Passes.h"
#include "mlir/Transforms/Utils.h"
using namespace mlir;
using namespace mlir::scf;
namespace {
struct SCFToStandardPass : public SCFToStandardBase<SCFToStandardPass> {
void runOnOperation() override;
};
// Create a CFG subgraph for the loop around its body blocks (if the body
// contained other loops, they have been already lowered to a flow of blocks).
// Maintain the invariants that a CFG subgraph created for any loop has a single
// entry and a single exit, and that the entry/exit blocks are respectively
// first/last blocks in the parent region. The original loop operation is
// replaced by the initialization operations that set up the initial value of
// the loop induction variable (%iv) and computes the loop bounds that are loop-
// invariant for affine loops. The operations following the original scf.for
// are split out into a separate continuation (exit) block. A condition block is
// created before the continuation block. It checks the exit condition of the
// loop and branches either to the continuation block, or to the first block of
// the body. The condition block takes as arguments the values of the induction
// variable followed by loop-carried values. Since it dominates both the body
// blocks and the continuation block, loop-carried values are visible in all of
// those blocks. Induction variable modification is appended to the last block
// of the body (which is the exit block from the body subgraph thanks to the
// invariant we maintain) along with a branch that loops back to the condition
// block. Loop-carried values are the loop terminator operands, which are
// forwarded to the branch.
//
// +---------------------------------+
// | <code before the ForOp> |
// | <definitions of %init...> |
// | <compute initial %iv value> |
// | br cond(%iv, %init...) |
// +---------------------------------+
// |
// -------| |
// | v v
// | +--------------------------------+
// | | cond(%iv, %init...): |
// | | <compare %iv to upper bound> |
// | | cond_br %r, body, end |
// | +--------------------------------+
// | | |
// | | -------------|
// | v |
// | +--------------------------------+ |
// | | body-first: | |
// | | <%init visible by dominance> | |
// | | <body contents> | |
// | +--------------------------------+ |
// | | |
// | ... |
// | | |
// | +--------------------------------+ |
// | | body-last: | |
// | | <body contents> | |
// | | <operands of yield = %yields>| |
// | | %new_iv =<add step to %iv> | |
// | | br cond(%new_iv, %yields) | |
// | +--------------------------------+ |
// | | |
// |----------- |--------------------
// v
// +--------------------------------+
// | end: |
// | <code after the ForOp> |
// | <%init visible by dominance> |
// +--------------------------------+
//
struct ForLowering : public OpRewritePattern<ForOp> {
using OpRewritePattern<ForOp>::OpRewritePattern;
LogicalResult matchAndRewrite(ForOp forOp,
PatternRewriter &rewriter) const override;
};
// Create a CFG subgraph for the scf.if operation (including its "then" and
// optional "else" operation blocks). We maintain the invariants that the
// subgraph has a single entry and a single exit point, and that the entry/exit
// blocks are respectively the first/last block of the enclosing region. The
// operations following the scf.if are split into a continuation (subgraph
// exit) block. The condition is lowered to a chain of blocks that implement the
// short-circuit scheme. The "scf.if" operation is replaced with a conditional
// branch to either the first block of the "then" region, or to the first block
// of the "else" region. In these blocks, "scf.yield" is unconditional branches
// to the post-dominating block. When the "scf.if" does not return values, the
// post-dominating block is the same as the continuation block. When it returns
// values, the post-dominating block is a new block with arguments that
// correspond to the values returned by the "scf.if" that unconditionally
// branches to the continuation block. This allows block arguments to dominate
// any uses of the hitherto "scf.if" results that they replaced. (Inserting a
// new block allows us to avoid modifying the argument list of an existing
// block, which is illegal in a conversion pattern). When the "else" region is
// empty, which is only allowed for "scf.if"s that don't return values, the
// condition branches directly to the continuation block.
//
// CFG for a scf.if with else and without results.
//
// +--------------------------------+
// | <code before the IfOp> |
// | cond_br %cond, %then, %else |
// +--------------------------------+
// | |
// | --------------|
// v |
// +--------------------------------+ |
// | then: | |
// | <then contents> | |
// | br continue | |
// +--------------------------------+ |
// | |
// |---------- |-------------
// | V
// | +--------------------------------+
// | | else: |
// | | <else contents> |
// | | br continue |
// | +--------------------------------+
// | |
// ------| |
// v v
// +--------------------------------+
// | continue: |
// | <code after the IfOp> |
// +--------------------------------+
//
// CFG for a scf.if with results.
//
// +--------------------------------+
// | <code before the IfOp> |
// | cond_br %cond, %then, %else |
// +--------------------------------+
// | |
// | --------------|
// v |
// +--------------------------------+ |
// | then: | |
// | <then contents> | |
// | br dom(%args...) | |
// +--------------------------------+ |
// | |
// |---------- |-------------
// | V
// | +--------------------------------+
// | | else: |
// | | <else contents> |
// | | br dom(%args...) |
// | +--------------------------------+
// | |
// ------| |
// v v
// +--------------------------------+
// | dom(%args...): |
// | br continue |
// +--------------------------------+
// |
// v
// +--------------------------------+
// | continue: |
// | <code after the IfOp> |
// +--------------------------------+
//
struct IfLowering : public OpRewritePattern<IfOp> {
using OpRewritePattern<IfOp>::OpRewritePattern;
LogicalResult matchAndRewrite(IfOp ifOp,
PatternRewriter &rewriter) const override;
};
struct ParallelLowering : public OpRewritePattern<mlir::scf::ParallelOp> {
using OpRewritePattern<mlir::scf::ParallelOp>::OpRewritePattern;
LogicalResult matchAndRewrite(mlir::scf::ParallelOp parallelOp,
PatternRewriter &rewriter) const override;
};
/// Create a CFG subgraph for this loop construct. The regions of the loop need
/// not be a single block anymore (for example, if other SCF constructs that
/// they contain have been already converted to CFG), but need to be single-exit
/// from the last block of each region. The operations following the original
/// WhileOp are split into a new continuation block. Both regions of the WhileOp
/// are inlined, and their terminators are rewritten to organize the control
/// flow implementing the loop as follows.
///
/// +---------------------------------+
/// | <code before the WhileOp> |
/// | br ^before(%operands...) |
/// +---------------------------------+
/// |
/// -------| |
/// | v v
/// | +--------------------------------+
/// | | ^before(%bargs...): |
/// | | %vals... = <some payload> |
/// | +--------------------------------+
/// | |
/// | ...
/// | |
/// | +--------------------------------+
/// | | ^before-last:
/// | | %cond = <compute condition> |
/// | | cond_br %cond, |
/// | | ^after(%vals...), ^cont |
/// | +--------------------------------+
/// | | |
/// | | -------------|
/// | v |
/// | +--------------------------------+ |
/// | | ^after(%aargs...): | |
/// | | <body contents> | |
/// | +--------------------------------+ |
/// | | |
/// | ... |
/// | | |
/// | +--------------------------------+ |
/// | | ^after-last: | |
/// | | %yields... = <some payload> | |
/// | | br ^before(%yields...) | |
/// | +--------------------------------+ |
/// | | |
/// |----------- |--------------------
/// v
/// +--------------------------------+
/// | ^cont: |
/// | <code after the WhileOp> |
/// | <%vals from 'before' region |
/// | visible by dominance> |
/// +--------------------------------+
///
/// Values are communicated between ex-regions (the groups of blocks that used
/// to form a region before inlining) through block arguments of their
/// entry blocks, which are visible in all other dominated blocks. Similarly,
/// the results of the WhileOp are defined in the 'before' region, which is
/// required to have a single existing block, and are therefore accessible in
/// the continuation block due to dominance.
struct WhileLowering : public OpRewritePattern<WhileOp> {
using OpRewritePattern<WhileOp>::OpRewritePattern;
LogicalResult matchAndRewrite(WhileOp whileOp,
PatternRewriter &rewriter) const override;
};
/// Optimized version of the above for the case of the "after" region merely
/// forwarding its arguments back to the "before" region (i.e., a "do-while"
/// loop). This avoid inlining the "after" region completely and branches back
/// to the "before" entry instead.
struct DoWhileLowering : public OpRewritePattern<WhileOp> {
using OpRewritePattern<WhileOp>::OpRewritePattern;
LogicalResult matchAndRewrite(WhileOp whileOp,
PatternRewriter &rewriter) const override;
};
} // namespace
LogicalResult ForLowering::matchAndRewrite(ForOp forOp,
PatternRewriter &rewriter) const {
Location loc = forOp.getLoc();
// Start by splitting the block containing the 'scf.for' into two parts.
// The part before will get the init code, the part after will be the end
// point.
auto *initBlock = rewriter.getInsertionBlock();
auto initPosition = rewriter.getInsertionPoint();
auto *endBlock = rewriter.splitBlock(initBlock, initPosition);
// Use the first block of the loop body as the condition block since it is the
// block that has the induction variable and loop-carried values as arguments.
// Split out all operations from the first block into a new block. Move all
// body blocks from the loop body region to the region containing the loop.
auto *conditionBlock = &forOp.region().front();
auto *firstBodyBlock =
rewriter.splitBlock(conditionBlock, conditionBlock->begin());
auto *lastBodyBlock = &forOp.region().back();
rewriter.inlineRegionBefore(forOp.region(), endBlock);
auto iv = conditionBlock->getArgument(0);
// Append the induction variable stepping logic to the last body block and
// branch back to the condition block. Loop-carried values are taken from
// operands of the loop terminator.
Operation *terminator = lastBodyBlock->getTerminator();
rewriter.setInsertionPointToEnd(lastBodyBlock);
auto step = forOp.step();
auto stepped = rewriter.create<AddIOp>(loc, iv, step).getResult();
if (!stepped)
return failure();
SmallVector<Value, 8> loopCarried;
loopCarried.push_back(stepped);
loopCarried.append(terminator->operand_begin(), terminator->operand_end());
rewriter.create<BranchOp>(loc, conditionBlock, loopCarried);
rewriter.eraseOp(terminator);
// Compute loop bounds before branching to the condition.
rewriter.setInsertionPointToEnd(initBlock);
Value lowerBound = forOp.lowerBound();
Value upperBound = forOp.upperBound();
if (!lowerBound || !upperBound)
return failure();
// The initial values of loop-carried values is obtained from the operands
// of the loop operation.
SmallVector<Value, 8> destOperands;
destOperands.push_back(lowerBound);
auto iterOperands = forOp.getIterOperands();
destOperands.append(iterOperands.begin(), iterOperands.end());
rewriter.create<BranchOp>(loc, conditionBlock, destOperands);
// With the body block done, we can fill in the condition block.
rewriter.setInsertionPointToEnd(conditionBlock);
auto comparison =
rewriter.create<CmpIOp>(loc, CmpIPredicate::slt, iv, upperBound);
rewriter.create<CondBranchOp>(loc, comparison, firstBodyBlock,
ArrayRef<Value>(), endBlock, ArrayRef<Value>());
// The result of the loop operation is the values of the condition block
// arguments except the induction variable on the last iteration.
rewriter.replaceOp(forOp, conditionBlock->getArguments().drop_front());
return success();
}
LogicalResult IfLowering::matchAndRewrite(IfOp ifOp,
PatternRewriter &rewriter) const {
auto loc = ifOp.getLoc();
// Start by splitting the block containing the 'scf.if' into two parts.
// The part before will contain the condition, the part after will be the
// continuation point.
auto *condBlock = rewriter.getInsertionBlock();
auto opPosition = rewriter.getInsertionPoint();
auto *remainingOpsBlock = rewriter.splitBlock(condBlock, opPosition);
Block *continueBlock;
if (ifOp.getNumResults() == 0) {
continueBlock = remainingOpsBlock;
} else {
continueBlock =
rewriter.createBlock(remainingOpsBlock, ifOp.getResultTypes());
rewriter.create<BranchOp>(loc, remainingOpsBlock);
}
// Move blocks from the "then" region to the region containing 'scf.if',
// place it before the continuation block, and branch to it.
auto &thenRegion = ifOp.thenRegion();
auto *thenBlock = &thenRegion.front();
Operation *thenTerminator = thenRegion.back().getTerminator();
ValueRange thenTerminatorOperands = thenTerminator->getOperands();
rewriter.setInsertionPointToEnd(&thenRegion.back());
rewriter.create<BranchOp>(loc, continueBlock, thenTerminatorOperands);
rewriter.eraseOp(thenTerminator);
rewriter.inlineRegionBefore(thenRegion, continueBlock);
// Move blocks from the "else" region (if present) to the region containing
// 'scf.if', place it before the continuation block and branch to it. It
// will be placed after the "then" regions.
auto *elseBlock = continueBlock;
auto &elseRegion = ifOp.elseRegion();
if (!elseRegion.empty()) {
elseBlock = &elseRegion.front();
Operation *elseTerminator = elseRegion.back().getTerminator();
ValueRange elseTerminatorOperands = elseTerminator->getOperands();
rewriter.setInsertionPointToEnd(&elseRegion.back());
rewriter.create<BranchOp>(loc, continueBlock, elseTerminatorOperands);
rewriter.eraseOp(elseTerminator);
rewriter.inlineRegionBefore(elseRegion, continueBlock);
}
rewriter.setInsertionPointToEnd(condBlock);
rewriter.create<CondBranchOp>(loc, ifOp.condition(), thenBlock,
/*trueArgs=*/ArrayRef<Value>(), elseBlock,
/*falseArgs=*/ArrayRef<Value>());
// Ok, we're done!
rewriter.replaceOp(ifOp, continueBlock->getArguments());
return success();
}
LogicalResult
ParallelLowering::matchAndRewrite(ParallelOp parallelOp,
PatternRewriter &rewriter) const {
Location loc = parallelOp.getLoc();
// For a parallel loop, we essentially need to create an n-dimensional loop
// nest. We do this by translating to scf.for ops and have those lowered in
// a further rewrite. If a parallel loop contains reductions (and thus returns
// values), forward the initial values for the reductions down the loop
// hierarchy and bubble up the results by modifying the "yield" terminator.
SmallVector<Value, 4> iterArgs = llvm::to_vector<4>(parallelOp.initVals());
SmallVector<Value, 4> ivs;
ivs.reserve(parallelOp.getNumLoops());
bool first = true;
SmallVector<Value, 4> loopResults(iterArgs);
for (auto loop_operands :
llvm::zip(parallelOp.getInductionVars(), parallelOp.lowerBound(),
parallelOp.upperBound(), parallelOp.step())) {
Value iv, lower, upper, step;
std::tie(iv, lower, upper, step) = loop_operands;
ForOp forOp = rewriter.create<ForOp>(loc, lower, upper, step, iterArgs);
ivs.push_back(forOp.getInductionVar());
auto iterRange = forOp.getRegionIterArgs();
iterArgs.assign(iterRange.begin(), iterRange.end());
if (first) {
// Store the results of the outermost loop that will be used to replace
// the results of the parallel loop when it is fully rewritten.
loopResults.assign(forOp.result_begin(), forOp.result_end());
first = false;
} else if (!forOp.getResults().empty()) {
// A loop is constructed with an empty "yield" terminator if there are
// no results.
rewriter.setInsertionPointToEnd(rewriter.getInsertionBlock());
rewriter.create<scf::YieldOp>(loc, forOp.getResults());
}
rewriter.setInsertionPointToStart(forOp.getBody());
}
// First, merge reduction blocks into the main region.
SmallVector<Value, 4> yieldOperands;
yieldOperands.reserve(parallelOp.getNumResults());
for (auto &op : *parallelOp.getBody()) {
auto reduce = dyn_cast<ReduceOp>(op);
if (!reduce)
continue;
Block &reduceBlock = reduce.reductionOperator().front();
Value arg = iterArgs[yieldOperands.size()];
yieldOperands.push_back(reduceBlock.getTerminator()->getOperand(0));
rewriter.eraseOp(reduceBlock.getTerminator());
rewriter.mergeBlockBefore(&reduceBlock, &op, {arg, reduce.operand()});
rewriter.eraseOp(reduce);
}
// Then merge the loop body without the terminator.
rewriter.eraseOp(parallelOp.getBody()->getTerminator());
Block *newBody = rewriter.getInsertionBlock();
if (newBody->empty())
rewriter.mergeBlocks(parallelOp.getBody(), newBody, ivs);
else
rewriter.mergeBlockBefore(parallelOp.getBody(), newBody->getTerminator(),
ivs);
// Finally, create the terminator if required (for loops with no results, it
// has been already created in loop construction).
if (!yieldOperands.empty()) {
rewriter.setInsertionPointToEnd(rewriter.getInsertionBlock());
rewriter.create<scf::YieldOp>(loc, yieldOperands);
}
rewriter.replaceOp(parallelOp, loopResults);
return success();
}
LogicalResult WhileLowering::matchAndRewrite(WhileOp whileOp,
PatternRewriter &rewriter) const {
OpBuilder::InsertionGuard guard(rewriter);
Location loc = whileOp.getLoc();
// Split the current block before the WhileOp to create the inlining point.
Block *currentBlock = rewriter.getInsertionBlock();
Block *continuation =
rewriter.splitBlock(currentBlock, rewriter.getInsertionPoint());
// Inline both regions.
Block *after = &whileOp.after().front();
Block *afterLast = &whileOp.after().back();
Block *before = &whileOp.before().front();
Block *beforeLast = &whileOp.before().back();
rewriter.inlineRegionBefore(whileOp.after(), continuation);
rewriter.inlineRegionBefore(whileOp.before(), after);
// Branch to the "before" region.
rewriter.setInsertionPointToEnd(currentBlock);
rewriter.create<BranchOp>(loc, before, whileOp.inits());
// Replace terminators with branches. Assuming bodies are SESE, which holds
// given only the patterns from this file, we only need to look at the last
// block. This should be reconsidered if we allow break/continue in SCF.
rewriter.setInsertionPointToEnd(beforeLast);
auto condOp = cast<ConditionOp>(beforeLast->getTerminator());
rewriter.replaceOpWithNewOp<CondBranchOp>(condOp, condOp.condition(), after,
condOp.args(), continuation,
ValueRange());
rewriter.setInsertionPointToEnd(afterLast);
auto yieldOp = cast<scf::YieldOp>(afterLast->getTerminator());
rewriter.replaceOpWithNewOp<BranchOp>(yieldOp, before, yieldOp.results());
// Replace the op with values "yielded" from the "before" region, which are
// visible by dominance.
rewriter.replaceOp(whileOp, condOp.args());
return success();
}
LogicalResult
DoWhileLowering::matchAndRewrite(WhileOp whileOp,
PatternRewriter &rewriter) const {
if (!llvm::hasSingleElement(whileOp.after()))
return rewriter.notifyMatchFailure(whileOp,
"do-while simplification applicable to "
"single-block 'after' region only");
Block &afterBlock = whileOp.after().front();
if (!llvm::hasSingleElement(afterBlock))
return rewriter.notifyMatchFailure(whileOp,
"do-while simplification applicable "
"only if 'after' region has no payload");
auto yield = dyn_cast<scf::YieldOp>(&afterBlock.front());
if (!yield || yield.results() != afterBlock.getArguments())
return rewriter.notifyMatchFailure(whileOp,
"do-while simplification applicable "
"only to forwarding 'after' regions");
// Split the current block before the WhileOp to create the inlining point.
OpBuilder::InsertionGuard guard(rewriter);
Block *currentBlock = rewriter.getInsertionBlock();
Block *continuation =
rewriter.splitBlock(currentBlock, rewriter.getInsertionPoint());
// Only the "before" region should be inlined.
Block *before = &whileOp.before().front();
Block *beforeLast = &whileOp.before().back();
rewriter.inlineRegionBefore(whileOp.before(), continuation);
// Branch to the "before" region.
rewriter.setInsertionPointToEnd(currentBlock);
rewriter.create<BranchOp>(whileOp.getLoc(), before, whileOp.inits());
// Loop around the "before" region based on condition.
rewriter.setInsertionPointToEnd(beforeLast);
auto condOp = cast<ConditionOp>(beforeLast->getTerminator());
rewriter.replaceOpWithNewOp<CondBranchOp>(condOp, condOp.condition(), before,
condOp.args(), continuation,
ValueRange());
// Replace the op with values "yielded" from the "before" region, which are
// visible by dominance.
rewriter.replaceOp(whileOp, condOp.args());
return success();
}
void mlir::populateLoopToStdConversionPatterns(
OwningRewritePatternList &patterns, MLIRContext *ctx) {
patterns.insert<ForLowering, IfLowering, ParallelLowering, WhileLowering>(
ctx);
patterns.insert<DoWhileLowering>(ctx, /*benefit=*/2);
}
void SCFToStandardPass::runOnOperation() {
OwningRewritePatternList patterns;
populateLoopToStdConversionPatterns(patterns, &getContext());
// Configure conversion to lower out scf.for, scf.if, scf.parallel and
// scf.while. Anything else is fine.
ConversionTarget target(getContext());
target.addIllegalOp<scf::ForOp, scf::IfOp, scf::ParallelOp, scf::WhileOp>();
target.markUnknownOpDynamicallyLegal([](Operation *) { return true; });
if (failed(
applyPartialConversion(getOperation(), target, std::move(patterns))))
signalPassFailure();
}
std::unique_ptr<Pass> mlir::createLowerToCFGPass() {
return std::make_unique<SCFToStandardPass>();
}
Reference in New Issue View Git Blame Copy Permalink